Re: RAID 5+1 and BAARF

Since I've done a - but BAARF often doesn't apply - reply today which still contains my position, here's a response.

On 10/30/06, ryan_gaffuri_at_comcast.net <ryan_gaffuri_at_comcast.net> wrote:
>
> I am more on the development side, so I have very little experience with
> hardware. I have read the articles on BAARF. I directly our infrastructure
> guy to the webpage and he is looking at it. Here is an email I got from him.
> Is there ever a cause to use RAID 5?
>

Yes. When your IO throughput isn't the problem or likely to become it. RAID5 is cheaper - though see below.

What % of the SAN is devoted to your critical databases? If all of it then the above - might just possibly be true - but I doubt it. It rather suggests that using RAID10 would imply 3 times as many disks/controllers etc, I'd be asking for a breakdown of how the cost is worked out.

I worked on this two years ago, so the information below is 2 years old.
>
> In SAN (EMC or Hitachi arrays) the issues pointed out by Milsap and
> others are avoided as follows, keeping the cost low:
>
> - There are spare disks on the array. The array takes care of
> automatically replacing a bad disk in the stripe set thus keeping the
> "reduced performance duration", that to a minimal.
>

What happens when 2 disks fail at once? CF what happens in a RAID 0+1 environment.

• With lot of cache, fiber backplane and parallel writes to stripes, the
  > "four I/O operations per read/write" issue is mitigated.
  >
  

Mitigated, yes, but not eliminated. At some point the cache will be exhausted, and then performance will hit the floor. Bear in mind that you will likely have Oracle and filesystem cache's involved as well so the caching algorithm of the hardware will have to be one that doesn't cache blocks held in the other two cache's - clever stuff hey, ask how the SAN does it. You'll also need the SAN to guarantee that all writes to CACHE will never be lost, does it?

-- 
Niall Litchfield
Oracle DBA
http://www.orawin.info

--
http://www.freelists.org/webpage/oracle-l